← Back to list

AI and Film: How I create a Shortfilm from scratch. Step 2

Finished script  Earlier I described how I started writing a script for my short film. This is my first work in such a plan but what’s…

Renas · 2025-02-25 08:27 · 4 claps · 6.6 min read
#ai #video-generation #chatgpt-vs-deepseek #short-film
Open on Medium ↗
Wiki topics: LLM · Large Language Models MM · Multimodal & Generative Media AI · AI · General 🎬 · Film & Television

AI and Film: How I create a Shortfilm from scratch. Step 2

The story of how I’m creating a movie using a set of AI tools instead of a film crew.

Photo by Mohamed Nohassi on Unsplash

Photo by Mohamed Nohassi on Unsplash

Finished script Earlier I described how I started writing a script for my short film. This is my first work in such a plan but what’s more interesting is that I want to do all visuals and voice-overs with AI.

Let’s go over the past topic a little bit. Well, I’m finally done with writing the script for the short film. If you’re interested in it I’ll attach it here later(just let me know about it). The experience turned out to be quite fun and entertaining, I revisited cult classics to refresh my memory and take a fresh look. Yes I also started writing reviews of the movies I’ve been watching, you can read them all on my profile.

In the beginning when I started writing about academic dramaturgy I didn’t know much, it was all based on my observation and understanding of logic, based on experience, so I decided to tighten up a bit on dramaturgy and delve a bit into stage craft and to my surprise I got it pretty much right. At one point I shared a little bit, one evening I was visiting my family and we decided to watch a movie directed by Leonid GaidaiIvan Vasilievich Changes Profession”, yes it is a Soviet movie, although I had only heard about it and seen a couple of scenes but never watched it, and after watching it I noticed that Gaidai used the same technique as me in his script (or I used the same technique as Gaidai). In the movie it was shown how a scientist in his experiments to create a time machine accidentally injured then his machine worked and he opened a portal to the time of Tsar Ivan the Terrible. At the end of the film it is shown that all this was his dream, and this is how Gaidai showed it, he divided the events by visual style, or rather when it was a dream, an illusion, he showed colorful color pictures and when the reality it was a usual dull black and white movie. In my work I also want to separate the illusory and the real world by the colors and visual forms, in reality everything is clear but dim and in illusion everything is bright and in some places distorted.

And that’s what I’m all about, that even with the lack of theory I compensated for it with the usual nazvidritnost, I watched quite a lot of different movies series mutfilms and anime, yes sometimes they were very specific and you sit through the whole movie in a complete misunderstanding, but still it’s a very great visual experience.

I also learned a lot about how to develop characters so that they seem very alive. ChatGPT and more recently DeepSeek have also played a big part in my knowledge. I think that if you write a lot of text this is an indispensable tool, but don’t think that chat will write instead of you, it’s just your assistant. You can use them to refresh your memory pretty quickly or find the information you need, or just analyze it.

Promt

Continuing the ChatGPT topic, I will say the following — it’s a wash. Yes in writing the script I have already started to communicate with the chat quite well, now less attempts go to that-so that it from the first time he understood what I need from him. Having compiled a small instruction I took on a long and painful writing promt, I will remind you that it is necessary to write a promt to each frame, and the script on the timing shows about 30 minutes of screen time, it is still to generate various frames that may be useful, in short, the task is not for one evening.

ChatGPT

ChatGPT

ChatGPT did a pretty good job, but you often need to edit its “worldview”. Since I mentioned DeepSeek, and since I’m currently observing a trend of confrontation, at this stage I’m still leaning towards the player in the green corner of the ring. Without word DeepSeek is doing great too, and his “DeepThink(R1)” mode is a pretty cool thing, but while it’s still pretty raw, ChatGPT is more flexible under the tag, which is both a plus and a minus on one hand. And this is either just me or in general, but in DeepSeek I very often encountered when it hangs, often at the first request. Yes, maybe in other areas it shows up better, but as an assistant in the letter so far not very much.

DeepSeek

DeepSeek

I also want to touch on the writing software that I use, yes I first started with the familiar Word and Pages, but rather moved on to other things, for writing articles and similar text I use Ulysses and for promts and various other information that I use as a notepad my choice stopped at Obsidian, quite a good thing, so sometimes I use it as an outline. For the script itself, it’s KIT Scenarist, which is just the right thing for a beginning writer. At the beginning I just needed a sheet to write on, but when your writing becomes a lot it is very important to structure it properly, and to be able to make a note for yourself in the future, so that in 2 weeks of pause when you open the project not to reread and not to remember everything that is going on here.

KIT Scenarist and Obsidian

KIT Scenarist and Obsidian

Video generation

So after the first written promts I tried video generation with HailuoAI. It’s a fascinating process to be honest, it’s not that feeling when you generate a couple of frames for a project and that’s it, you’re creating a whole story, and it’s very important that you don’t have differences in characters or locations throughout. When generating characters I started to make a general promt, first I described the scene and the character that were not changed in the next frames and then already prescribed the actions in the scene. Sometimes you have to redo or rewrite the promt 10 times in pursuit of the very frame that you want. So I had when I needed a close-up shot where the gushchina grabs the guy by the shoulder and pulls on himself, how many different styles and visual solutions I was offered, but all passed by, after many generations I still got a relatively necessary moment. Separately I want to say that at the moment neuronki on video generation come to a pretty decent level, many moments in the frames I did not describe, but the AI very accurately transferred the general style from frame to frame, also a lot of innovations in functionality, the fact that you can already do one button and do not prescribe the exact movement and camera angle so detailed as before it is definitely cool.

General promt

General promt

HailuoAI

HailuoAI

Music generation

I am without magnification was delighted. SUNO you guys are cool. Before that I used mainly photo and video generation, for sound used neural networks with narrow tasks, as very often it is “Adobe Podcast”, but my surprise was no limits when I first heard what generated. I realize that nowadays it’s not as rare as generating anything, just look at the level of top AI, but I was still excited as a kid. It’s pretty cool that everything is just a simple few words you get a full-fledged quite a quality song, not a melody or something like that, but a song, with lyrics and cool voices. A song, Carl! Today, nothing prevents you from making a YouTube channel, generate a song and a clip to it and collect millions of views. For 10 years we could only dream of such a thing.

[embed]Shadow City by @focusedprodigy4065 | Suno ambient, dark song. Listen and make your own with Suno.suno.com

[embed]Shadow Dance by @focusedprodigy4065 | Suno ambient, electronic, dark wave, world song. Listen and make your own with Suno.suno.com

https://suno.com/song/a5f2bc80–943c-4b6a-b071–06a1d6d502d7

Conclusions

Well and I couldn’t wait to edit at least the first seconds, yet I will attach below a segment that is more or less ready. I warn you that this is still very raw material. I will not go into the editing because only a couple of frames are made.

DaVinci Resolve

DaVinci Resolve

Well, dear readers, this is only a small part of everything that awaits me in the future, my next steps is to do all the generation of video series, then start generating dialog. But of course I get an incredible high and pleasure from my work, I also want you to also get such emotions from your favorite business and develop in it.

[embed]

https://medium.com/@renas_/list/about-making-your-own-movie-4ff64488fc6e


메타데이터
post_id
f7dd9e41b42e
slug
ai-and-film-how-i-create-a-shortfilm-from-scratch-step-2-f7dd9e41b42e
url
https://medium.com/@renas_/ai-and-film-how-i-create-a-shortfilm-from-scratch-step-2-f7dd9e41b42e
canonical_url
https://medium.com/@renas_/ai-and-film-how-i-create-a-shortfilm-from-scratch-step-2-f7dd9e41b42e
author_url
https://medium.com/@renas_
status
ok
fetched_at
2026-07-20 21:03:38